Hybrid Enhanced Universal Dependencies Parsing

What it is about

The system predicts enhanced UD in two steps: a graph-based parser builds the basic tree, then a set of linguistic rules adds the enhanced edges, using only the tree and the UPOS tags, so that the rules are language-independent. A classifier (ICSIBoost, over features such as the rule name, the language, the head UPOS and the distance to the head) decides whether each rule application is kept, and thresholds between 30 and 60 percent gave the best ELAS. The system ranked second in the IWPT 2020 shared task with an average ELAS of 82.60 and an EULAS of 84.6. On gold basic trees the rule layer alone scores at least 88.8 ELAS on every treebank in the paper's gold-tree table, so most of the loss comes from the parser.

Ideas

• I1. Accept-or-reject classifier per rule in the Rust layer: a small decision tree trained on EWT train with deterministic (BTreeMap) features decides whether each enhanced rule fires; measure ELAS on EWT dev over the thresholds 0.3 to 0.6, as the paper does.
• I2. Ceiling check first: run our enhanced layer on gold basic trees of EWT dev before any parser work; the paper's gold-tree numbers say the rule layer is not the bottleneck, so a gap there would be a bug to fix, not a model to tune.
• I3. Language-independence test: write the rules as UPOS-and-tree patterns in the en/seeds form, and check on GUM (test only) whether they transfer with no language-specific change.

For Mova

The enhanced layer would sit beside tools/treeagree (src/rules.rs, src/tree.rs) and use the parser from en/src/parse.rs or gparse. The chapter is 20. Syntax: the Universal Dependencies frame. Measure first the ELAS of the rule layer on gold basic trees of EWT dev, which gives the ceiling for any parser.

Topics: Universal Dependencies

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2020.iwpt-1.18

@inproceedings{heinecke2020hybrid,
  title = {{Hybrid Enhanced Universal Dependencies Parsing}},
  author = {Heinecke, Johannes},
  booktitle = {Proceedings of the 16th International Conference on Parsing Technologies and the IWPT 2020 Shared Task on Parsing into Enhanced Universal Dependencies},
  year = {2020},
  doi = {10.18653/v1/2020.iwpt-1.18},
  url = {https://doi.org/10.18653/v1/2020.iwpt-1.18}
}

References

19 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Mikhail Arkhipov, Maria Trofimova, Yuri Kuratov, and Alexey Sorokin. 2019. Tuning Multilingual Transformers for Language-Specific Named Entity Recognition. In ACL 2019, pages 89–93, Florence. find
2. Language ar bg cs en et fi fr it lv lt nl pl ru sk se ta uk Average Tokens 99.97 99.89 99.85 99.22 99.70 99.69 99.54 99.90 99.34 99.94 99.71 99.33 99.48 99.99 99.72 99.49 99.79 99.68 Words 94.57 99.89 99.85 99.22 99.70 99.68 99.13 99.82 99.34 99.94 99.71 99.79 99.48 99.99 99.72 94.96 99.75 99.09 Sen… find our card
3. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grace, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Unsupervised Cross-lingual Representation Learning at Scale. https://arxiv.org/abs/1911.02116. ↗
4. Jacob Devlin, Ming-Wei Chang Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language under-standing. In NAACL, page 4171–4186, Minneapolis. Association for Computational Linguistics. find
5. Benoît Favre, Dilek Hakkani-Tür, Slav Petrov, and Klein Dan. 2008. Efficient sentence segmentation using syntactic features. In Spoken Language Technology Workshop, pages 77–80. IEEE. find
6. Joakim Nivre, Marie-Catherine de Marneffe, Filip Ginter, Yoav Goldberg, Yoav Goldberg, Jan Hajič, Manning Christopher D., Ryan McDonald, Slav Petrov, Sampo Pyysalo, Natalia Silveira, Reut Tsarfaty, and Daniel Zeman. 2016. Universal Dependencies v1: A Multilingual Treebank Collection. In the tenth i… find our card
7. Joakim Nivre, Paola Marongiu, Filip Ginter, Jenna Kanerva, Simonetta Montemagni, Sebastian Schuster, and Maria Simi. 2018. Enhancing Universal Dependency Treebanks: A Case Study. In Proceedings of the Second Workshop on Universal Dependencies (UDW 2018), pages 102–107, Brussels, Belgium. Association… find our card
8. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kendon Lee, and Luke Zettlemoyer. 2018. Deep Contextualized Word Representations. In NAACL, page 2227–2237, New Orleans, Louisiana. Association for Computational Linguistics. find
9. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. RoBERTa: A Robustly Optimized BERT Pretraining Approach. https://arxiv.org/abs/1907.11692. ↗
10. Sebastian Schuster and Christopher D. Manning. 2016. Enhanced English Universal Dependencies: An Improved Representation for Natural Language Understanding Tasks. In the tenth international conference on Language Resources and Evaluation, pages 2371–2378, Portorož, Slovenia. European Language Resour… find our card
11. Louis Martin, Benjamin Muller, Pedro Javier Ortiz Suárez, Yoann Dupont, Laurent Romary, Éric Villemonte de la Clergerie, Djamé Seddah, and Benoît Sagot. 2019. CamemBERT: a Tasty French Language Model. https://arxiv.org/abs/1911.03894. ↗
12. Sebastian Schuster, Joachim Nivre, and Christopher D. Manning. 2018. Sentences with Gapping: Parsing and Reconstructing Elided Predicates. In NAACL, pages 1156–1168, New Orleans, Louisiana. Association for Computational Linguistics. find our card
13. Joakim Nivre and Chiao-Ting Fang. 2017. Universal Dependency Evaluation. In Proceedings of the NoDaLiDa 2017 Workshop on Universal Dependencies, pages 86–95, Göteborg. find
14. Milan Straka. 2018. UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task. In Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 197–207, Brussels. Association for Computational Linguistics. find our card
15. Milan Straka and Jana Straková. 2017. Tokenizing, POS Tagging, Lemmatizing and Parsing UD 2.0 with UDPipe. In ACL 2017, pages 88–99, Vancouver. find our card
16. Antti Virtanen, Jenna Kanerva, Rami Ilo, Juhani Luotolahti, Tapio Salakoski, Filip Ginter, and Sampo Pyysalo. 2019. Multilingual is not enough: BERT for Finnish. https://arxiv.org/abs/1912.07076. ↗
17. Wietse de Vries, Andreas van Cranenburgh, Arianna Bisazza, Tommaso Caselli, Gertjan van Noord, and Malvina Nissim. 2019. BERTje: A Dutch BERT Model. https://arxiv.org/abs/1912.09582. ↗
18. Daniel Zeman, Jan Hajič, Martin Popel, Martin Potthast, Milan Straka, Filip Ginter, Joakim Nivre, and Slav Petrov. 2018. find
19. CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. In Proceedings of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 1–21, Brussels. Association for Computational Linguistics. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/ud-2020-heinecke-hybrid-enhanced-universal-dependencies-parsing/